]> git.saurik.com Git - apple/xnu.git/blobdiff - bsd/net/bpf.c
xnu-1699.26.8.tar.gz
[apple/xnu.git] / bsd / net / bpf.c
index 1da778677df12ae6e05ac8cea043e169f46e75f1..e370dfc5ed2d12bbb4c1aad57c0ebc14cc8f150f 100644 (file)
@@ -1,23 +1,29 @@
 /*
- * Copyright (c) 2000 Apple Computer, Inc. All rights reserved.
+ * Copyright (c) 2000-2010 Apple Inc. All rights reserved.
  *
- * @APPLE_LICENSE_HEADER_START@
+ * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
  * 
- * The contents of this file constitute Original Code as defined in and
- * are subject to the Apple Public Source License Version 1.1 (the
- * "License").  You may not use this file except in compliance with the
- * License.  Please obtain a copy of the License at
- * http://www.apple.com/publicsource and read it before using this file.
+ * This file contains Original Code and/or Modifications of Original Code
+ * as defined in and that are subject to the Apple Public Source License
+ * Version 2.0 (the 'License'). You may not use this file except in
+ * compliance with the License. The rights granted to you under the License
+ * may not be used to create, or enable the creation or redistribution of,
+ * unlawful or unlicensed copies of an Apple operating system, or to
+ * circumvent, violate, or enable the circumvention or violation of, any
+ * terms of an Apple operating system software license agreement.
  * 
- * This Original Code and all software distributed under the License are
- * distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY KIND, EITHER
+ * Please obtain a copy of the License at
+ * http://www.opensource.apple.com/apsl/ and read it before using this file.
+ * 
+ * The Original Code and all software distributed under the License are
+ * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE OR NON-INFRINGEMENT.  Please see the
- * License for the specific language governing rights and limitations
- * under the License.
+ * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
+ * Please see the License for the specific language governing rights and
+ * limitations under the License.
  * 
- * @APPLE_LICENSE_HEADER_END@
+ * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
  */
 /*
  * Copyright (c) 1990, 1991, 1993
  *
  * $FreeBSD: src/sys/net/bpf.c,v 1.59.2.5 2001/01/05 04:49:09 jdp Exp $
  */
+/*
+ * NOTICE: This file was modified by SPARTA, Inc. in 2005 to introduce
+ * support for mandatory and extensible security protections.  This notice
+ * is included in support of clause 2.2 (b) of the Apple Public License,
+ * Version 2.0.
+ */
 
 #include "bpf.h"
 
@@ -81,6 +93,9 @@
 #include <sys/sockio.h>
 #include <sys/ttycom.h>
 #include <sys/filedesc.h>
+#include <sys/uio_internal.h>
+#include <sys/file_internal.h>
+#include <sys/event.h>
 
 #if defined(sparc) && BSD < 199103
 #include <sys/stream.h>
 #include <sys/sysctl.h>
 #include <net/firewire.h>
 
-#include <machine/ansi.h>
 #include <miscfs/devfs/devfs.h>
 #include <net/dlil.h>
 
-#if NBPFILTER > 0
+#include <kern/locks.h>
+#include <kern/thread_call.h>
+
+#if CONFIG_MACF_NET
+#include <security/mac_framework.h>
+#endif /* MAC_NET */
+
+extern int tvtohz(struct timeval *);
 
 /*
  * Older BSDs don't have kernel malloc.
@@ -126,12 +147,15 @@ static caddr_t bpf_alloc();
 /*
  * The default read buffer size is patchable.
  */
-static int bpf_bufsize = BPF_BUFSIZE;
-SYSCTL_INT(_debug, OID_AUTO, bpf_bufsize, CTLFLAG_RW
+static unsigned int bpf_bufsize = BPF_BUFSIZE;
+SYSCTL_INT(_debug, OID_AUTO, bpf_bufsize, CTLFLAG_RW | CTLFLAG_LOCKED,
        &bpf_bufsize, 0, "");
-static int bpf_maxbufsize = BPF_MAXBUFSIZE;
-SYSCTL_INT(_debug, OID_AUTO, bpf_maxbufsize, CTLFLAG_RW
+__private_extern__ unsigned int bpf_maxbufsize = BPF_MAXBUFSIZE;
+SYSCTL_INT(_debug, OID_AUTO, bpf_maxbufsize, CTLFLAG_RW | CTLFLAG_LOCKED,
        &bpf_maxbufsize, 0, "");
+static unsigned int bpf_maxdevices = 256;
+SYSCTL_UINT(_debug, OID_AUTO, bpf_maxdevices, CTLFLAG_RW | CTLFLAG_LOCKED,
+       &bpf_maxdevices, 0, "");
 
 /*
  *  bpf_iflist is the list of interfaces; each corresponds to an ifnet
@@ -143,169 +167,199 @@ static struct bpf_if    *bpf_iflist;
  * BSD now stores the bpf_d in the dev_t which is a struct
  * on their system. Our dev_t is an int, so we still store
  * the bpf_d in a separate table indexed by minor device #.
+ *
+ * The value stored in bpf_dtab[n] represent three states:
+ *  0: device not opened
+ *  1: device opening or closing
+ *  other: device <n> opened with pointer to storage
  */
 static struct bpf_d    **bpf_dtab = NULL;
-static int                     bpf_dtab_size = 0;
-static int                     nbpfilter = 0;
+static unsigned int bpf_dtab_size = 0;
+static unsigned int    nbpfilter = 0;
+
+static lck_mtx_t               *bpf_mlock;
+static lck_grp_t               *bpf_mlock_grp;
+static lck_grp_attr_t  *bpf_mlock_grp_attr;
+static lck_attr_t              *bpf_mlock_attr;
 
 /*
  * Mark a descriptor free by making it point to itself.
  * This is probably cheaper than marking with a constant since
  * the address should be in a register anyway.
  */
-#define D_ISFREE(d) ((d) == (d)->bd_next)
-#define D_MARKFREE(d) ((d)->bd_next = (d))
-#define D_MARKUSED(d) ((d)->bd_next = 0)
 #endif /* __APPLE__ */
 
-static int     bpf_allocbufs __P((struct bpf_d *));
-static void    bpf_attachd __P((struct bpf_d *d, struct bpf_if *bp));
-static void    bpf_detachd __P((struct bpf_d *d));
-static void    bpf_freed __P((struct bpf_d *));
-static void    bpf_mcopy __P((const void *, void *, size_t));
-static int     bpf_movein __P((struct uio *, int,
-                   struct mbuf **, struct sockaddr *, int *));
-static int     bpf_setif __P((struct bpf_d *, struct ifreq *));
-static inline void
-               bpf_wakeup __P((struct bpf_d *));
-static void    catchpacket __P((struct bpf_d *, u_char *, u_int,
-                   u_int, void (*)(const void *, void *, size_t)));
-static void    reset_d __P((struct bpf_d *));
-static int      bpf_setf __P((struct bpf_d *, struct bpf_program *));
+static int     bpf_allocbufs(struct bpf_d *);
+static errno_t bpf_attachd(struct bpf_d *d, struct bpf_if *bp);
+static void    bpf_detachd(struct bpf_d *d);
+static void    bpf_freed(struct bpf_d *);
+static void    bpf_mcopy(const void *, void *, size_t);
+static int     bpf_movein(struct uio *, int,
+                   struct mbuf **, struct sockaddr *, int *);
+static int     bpf_setif(struct bpf_d *, ifnet_t ifp, u_int32_t dlt);
+static void bpf_timed_out(void *, void *);
+static void bpf_wakeup(struct bpf_d *);
+static void    catchpacket(struct bpf_d *, u_char *, u_int,
+                   u_int, void (*)(const void *, void *, size_t));
+static void    reset_d(struct bpf_d *);
+static int bpf_setf(struct bpf_d *, u_int bf_len, user_addr_t bf_insns);
+static int     bpf_getdltlist(struct bpf_d *, struct bpf_dltlist *,
+    struct proc *);
+static int     bpf_setdlt(struct bpf_d *, u_int);
 
 /*static  void *bpf_devfs_token[MAXBPFILTER];*/
 
 static  int bpf_devsw_installed;
 
-void bpf_init __P((void *unused));
-
+void bpf_init(void *unused);
+static int bpf_tap_callback(struct ifnet *ifp, struct mbuf *m);
 
 /*
  * Darwin differs from BSD here, the following are static
  * on BSD and not static on Darwin.
  */
-       d_open_t        bpfopen;
-       d_close_t       bpfclose;
-       d_read_t        bpfread;
-       d_write_t       bpfwrite;
-       d_ioctl_t       bpfioctl;
-       select_fcn_t    bpfpoll;
-
-#ifdef __APPLE__
-void bpf_mtap(struct ifnet *, struct mbuf *);
+       d_open_t            bpfopen;
+       d_close_t           bpfclose;
+       d_read_t            bpfread;
+       d_write_t           bpfwrite;
+    ioctl_fcn_t            bpfioctl;
+    select_fcn_t       bpfselect;
 
-int    bpfopen(), bpfclose(), bpfread(), bpfwrite(), bpfioctl(),
-               bpfpoll();
-#endif
 
 /* Darwin's cdevsw struct differs slightly from BSDs */
 #define CDEV_MAJOR 23
 static struct cdevsw bpf_cdevsw = {
-       /* open */      bpfopen,
-       /* close */     bpfclose,
-       /* read */      bpfread,
-       /* write */     bpfwrite,
-       /* ioctl */     bpfioctl,
-       /* stop */      nulldev,
-       /* reset */     nulldev,
-       /* tty */       NULL,
-       /* select */    bpfpoll,
-       /* mmap */      eno_mmap,
+       /* open */          bpfopen,
+       /* close */         bpfclose,
+       /* read */          bpfread,
+       /* write */         bpfwrite,
+       /* ioctl */         bpfioctl,
+       /* stop */              eno_stop,
+       /* reset */             eno_reset,
+       /* tty */               NULL,
+       /* select */    bpfselect,
+       /* mmap */              eno_mmap,
        /* strategy*/   eno_strat,
-       /* getc */      eno_getc,
-       /* putc */      eno_putc,
-       /* type */      0
+       /* getc */              eno_getc,
+       /* putc */              eno_putc,
+       /* type */              0
 };
 
 #define SOCKADDR_HDR_LEN          offsetof(struct sockaddr, sa_data)
 
 static int
-bpf_movein(uio, linktype, mp, sockp, datlen)
-       register struct uio *uio;
-       int linktype, *datlen;
-       register struct mbuf **mp;
-       register struct sockaddr *sockp;
+bpf_movein(struct uio *uio, int linktype, struct mbuf **mp, struct sockaddr *sockp, int *datlen)
 {
        struct mbuf *m;
        int error;
        int len;
+       uint8_t sa_family;
        int hlen;
 
-       /*
-        * Build a sockaddr based on the data link layer type.
-        * We do this at this level because the ethernet header
-        * is copied directly into the data field of the sockaddr.
-        * In the case of SLIP, there is no header and the packet
-        * is forwarded as is.
-        * Also, we are careful to leave room at the front of the mbuf
-        * for the link level header.
-        */
        switch (linktype) {
-
+       
+#if SLIP
        case DLT_SLIP:
-               sockp->sa_family = AF_INET;
+               sa_family = AF_INET;
                hlen = 0;
                break;
-
+#endif /* SLIP */
+       
        case DLT_EN10MB:
-               sockp->sa_family = AF_UNSPEC;
+               sa_family = AF_UNSPEC;
                /* XXX Would MAXLINKHDR be better? */
                hlen = sizeof(struct ether_header);
                break;
-
+       
+#if FDDI
        case DLT_FDDI:
-#if defined(__FreeBSD__) || defined(__bsdi__)
-               sockp->sa_family = AF_IMPLINK;
+       #if defined(__FreeBSD__) || defined(__bsdi__)
+               sa_family = AF_IMPLINK;
                hlen = 0;
-#else
-               sockp->sa_family = AF_UNSPEC;
+       #else
+               sa_family = AF_UNSPEC;
                /* XXX 4(FORMAC)+6(dst)+6(src)+3(LLC)+5(SNAP) */
                hlen = 24;
-#endif
+       #endif
                break;
-
+#endif /* FDDI */
+       
        case DLT_RAW:
        case DLT_NULL:
-               sockp->sa_family = AF_UNSPEC;
+               sa_family = AF_UNSPEC;
                hlen = 0;
                break;
-
-#ifdef __FreeBSD__
+       
+       #ifdef __FreeBSD__
        case DLT_ATM_RFC1483:
                /*
                 * en atm driver requires 4-byte atm pseudo header.
                 * though it isn't standard, vpi:vci needs to be
                 * specified anyway.
                 */
-               sockp->sa_family = AF_UNSPEC;
+               sa_family = AF_UNSPEC;
                hlen = 12;      /* XXX 4(ATM_PH) + 3(LLC) + 5(SNAP) */
                break;
-#endif
+       #endif
+
        case DLT_PPP:
-               sockp->sa_family = AF_UNSPEC;
+               sa_family = AF_UNSPEC;
                hlen = 4;       /* This should match PPP_HDRLEN */
                break;
-
+       
        case DLT_APPLE_IP_OVER_IEEE1394:
-               sockp->sa_family = AF_UNSPEC;
+               sa_family = AF_UNSPEC;
                hlen = sizeof(struct firewire_header);
                break;
 
+       case DLT_IEEE802_11:            /* IEEE 802.11 wireless */
+               sa_family = AF_IEEE80211;
+               hlen = 0;
+               break;
+       
+       case DLT_IEEE802_11_RADIO:
+               sa_family = AF_IEEE80211;
+               hlen = 0;
+               break;
+
        default:
                return (EIO);
        }
-       if ((hlen + SOCKADDR_HDR_LEN) > sockp->sa_len) {
-               return (EIO);
-       }
-       len = uio->uio_resid;
+
+       // LP64todo - fix this!
+       len = uio_resid(uio);
        *datlen = len - hlen;
        if ((unsigned)len > MCLBYTES)
                return (EIO);
 
+       if (sockp) {
+               /*
+                * Build a sockaddr based on the data link layer type.
+                * We do this at this level because the ethernet header
+                * is copied directly into the data field of the sockaddr.
+                * In the case of SLIP, there is no header and the packet
+                * is forwarded as is.
+                * Also, we are careful to leave room at the front of the mbuf
+                * for the link level header.
+                */
+               if ((hlen + SOCKADDR_HDR_LEN) > sockp->sa_len) {
+                       return (EIO);
+               }
+               sockp->sa_family = sa_family;
+       } else {
+               /*
+                * We're directly sending the packet data supplied by
+                * the user; we don't need to make room for the link
+                * header, and don't need the header length value any
+                * more, so set it to 0.
+                */
+               hlen = 0;
+       }
+       
        MGETHDR(m, M_WAIT, MT_DATA);
        if (m == 0)
                return (ENOBUFS);
-       if (len > MHLEN) {
+       if ((unsigned)len > MHLEN) {
 #if BSD >= 199103
                MCLGET(m, M_WAIT);
                if ((m->m_flags & M_EXT) == 0) {
@@ -320,6 +374,7 @@ bpf_movein(uio, linktype, mp, sockp, datlen)
        m->m_pkthdr.len = m->m_len = len;
        m->m_pkthdr.rcvif = NULL;
        *mp = m;
+       
        /*
         * Make room for link header.
         */
@@ -336,100 +391,99 @@ bpf_movein(uio, linktype, mp, sockp, datlen)
                        goto bad;
        }
        error = UIOMOVE(mtod(m, caddr_t), len - hlen, UIO_WRITE, uio);
-       if (!error)
-               return (0);
+       if (error)
+               goto bad;
+       
+       /* Check for multicast destination */
+       switch (linktype) {
+               case DLT_EN10MB: {
+                       struct ether_header *eh = mtod(m, struct ether_header *);
+                       
+                       if (ETHER_IS_MULTICAST(eh->ether_dhost)) {
+                               if (_ether_cmp(etherbroadcastaddr, eh->ether_dhost) == 0)
+                                       m->m_flags |= M_BCAST;
+                               else
+                                       m->m_flags |= M_MCAST;
+                       }
+                       break;
+               }
+       }
+       
+       return 0;
  bad:
        m_freem(m);
        return (error);
 }
 
 #ifdef __APPLE__
-/* Callback registered with Ethernet driver. */
-int bpf_tap_callback(struct ifnet *ifp, struct mbuf *m)
-{
-    boolean_t funnel_state;
-
-    funnel_state = thread_funnel_set(network_flock, TRUE);
-     
-    /*
-     * Do nothing if the BPF tap has been turned off.
-     * This is to protect from a potential race where this
-     * call blocks on the funnel lock. And in the meantime
-     * BPF is turned off, which will clear if_bpf.
-     */
-    if (ifp->if_bpf)
-        bpf_mtap(ifp, m);
-
-    thread_funnel_set(network_flock, funnel_state);
-    return 0;
-}
 
 /*
- * Returns 1 on sucess, 0 on failure
+ * The dynamic addition of a new device node must block all processes that are opening 
+ * the last device so that no process will get an unexpected ENOENT 
  */
-static int
-bpf_dtab_grow(int increment)
-{
-       struct bpf_d **new_dtab = NULL;
-       
-       new_dtab = (struct bpf_d **)_MALLOC(sizeof(struct bpf_d *) * (bpf_dtab_size + increment), M_DEVBUF, M_WAIT);
-       if  (new_dtab == NULL)
-               return 0;
-
-       if (bpf_dtab) {
-               struct bpf_d **old_dtab;
-
-               bcopy(bpf_dtab, new_dtab, sizeof(struct bpf_d *) * bpf_dtab_size);
-               /*
-                * replace must be atomic with respect to free do bpf_dtab
-                * is always valid.
-                */
-               old_dtab = bpf_dtab;
-       bpf_dtab = new_dtab;
-               _FREE(old_dtab, M_DEVBUF);
-       } 
-       else    bpf_dtab = new_dtab;
-       
-       bzero(bpf_dtab + bpf_dtab_size, sizeof(struct bpf_d *) * increment);
-       
-       bpf_dtab_size += increment;
-       
-       return 1;
-}
-
-static struct bpf_d *
+static void
 bpf_make_dev_t(int maj)
 {
-       struct bpf_d *d;
-       
-       if (nbpfilter >= bpf_dtab_size && bpf_dtab_grow(NBPFILTER) == 0)
-               return NULL;
+       static int              bpf_growing = 0;
+       unsigned int    cur_size = nbpfilter, i;
 
-       d = (struct bpf_d *)_MALLOC(sizeof(struct bpf_d), M_DEVBUF, M_WAIT);
-       if (d != NULL) {
-               int i = nbpfilter++;
+       if (nbpfilter >= bpf_maxdevices)
+               return;
 
-               bzero(d, sizeof(struct bpf_d));
-               bpf_dtab[i] = d;
-               D_MARKFREE(bpf_dtab[i]);
-               /*bpf_devfs_token[i]  = */devfs_make_node(makedev(maj, i),
-                                                       DEVFS_CHAR, UID_ROOT, GID_WHEEL, 0600,
-                                                       "bpf%d", i);
+       while (bpf_growing) {
+               /* Wait until new device has been created */
+               (void)tsleep((caddr_t)&bpf_growing, PZERO, "bpf_growing", 0);
+       }
+       if (nbpfilter > cur_size) {
+               /* other thread grew it already */
+               return;
        }
-       return d;
+       bpf_growing = 1;
+       
+       /* need to grow bpf_dtab first */
+       if (nbpfilter == bpf_dtab_size) {
+               int new_dtab_size;
+               struct bpf_d **new_dtab = NULL;
+               struct bpf_d **old_dtab = NULL;
+               
+               new_dtab_size = bpf_dtab_size + NBPFILTER;      
+               new_dtab = (struct bpf_d **)_MALLOC(sizeof(struct bpf_d *) * new_dtab_size, M_DEVBUF, M_WAIT);
+               if (new_dtab == 0) {
+                       printf("bpf_make_dev_t: malloc bpf_dtab failed\n");
+                       goto done;
+               }
+               if (bpf_dtab) {
+                       bcopy(bpf_dtab, new_dtab, 
+                                 sizeof(struct bpf_d *) * bpf_dtab_size);
+               }
+               bzero(new_dtab + bpf_dtab_size, 
+                         sizeof(struct bpf_d *) * NBPFILTER);
+               old_dtab = bpf_dtab;
+               bpf_dtab = new_dtab;
+               bpf_dtab_size = new_dtab_size;
+               if (old_dtab != NULL)
+                       _FREE(old_dtab, M_DEVBUF);
+       }
+       i = nbpfilter++;
+       (void) devfs_make_node(makedev(maj, i),
+                               DEVFS_CHAR, UID_ROOT, GID_WHEEL, 0600,
+                               "bpf%d", i);
+done:
+       bpf_growing = 0;
+       wakeup((caddr_t)&bpf_growing);
 }
 
 #endif
 
 /*
  * Attach file to the bpf interface, i.e. make d listen on bp.
- * Must be called at splimp.
  */
-static void
-bpf_attachd(d, bp)
-       struct bpf_d *d;
-       struct bpf_if *bp;
+static errno_t
+bpf_attachd(struct bpf_d *d, struct bpf_if *bp)
 {
+       int first = bp->bif_dlist == NULL;
+       int     error = 0;
+       
        /*
         * Point d at bp, and add d to the interface's list of listeners.
         * Finally, point the driver's bpf cookie at the interface so
@@ -438,125 +492,220 @@ bpf_attachd(d, bp)
        d->bd_bif = bp;
        d->bd_next = bp->bif_dlist;
        bp->bif_dlist = d;
+       
+       if (first) {
+               /* Find the default bpf entry for this ifp */
+               if (bp->bif_ifp->if_bpf == NULL) {
+                       struct bpf_if   *primary;
+                       
+                       for (primary = bpf_iflist; primary && primary->bif_ifp != bp->bif_ifp;
+                                primary = primary->bif_next)
+                               ;
+               
+                       bp->bif_ifp->if_bpf = primary;
+               }
+               
+               /* Only call dlil_set_bpf_tap for primary dlt */
+               if (bp->bif_ifp->if_bpf == bp)
+                       dlil_set_bpf_tap(bp->bif_ifp, BPF_TAP_INPUT_OUTPUT, bpf_tap_callback);
+               
+               if (bp->bif_tap)
+                       error = bp->bif_tap(bp->bif_ifp, bp->bif_dlt, BPF_TAP_INPUT_OUTPUT);
+       }
 
-       bp->bif_ifp->if_bpf = bp;
-
-#ifdef __APPLE__
-       if (bp->bif_ifp->if_set_bpf_tap)
-               (*bp->bif_ifp->if_set_bpf_tap)(bp->bif_ifp, BPF_TAP_INPUT_OUTPUT, bpf_tap_callback);
-#endif
+       return error;
 }
 
 /*
  * Detach a file from its interface.
  */
 static void
-bpf_detachd(d)
-       struct bpf_d *d;
+bpf_detachd(struct bpf_d *d)
 {
        struct bpf_d **p;
        struct bpf_if *bp;
-#ifdef __APPLE__
        struct ifnet  *ifp;
 
        ifp = d->bd_bif->bif_ifp;
-
-#endif
-
        bp = d->bd_bif;
+       
+       /* Remove d from the interface's descriptor list. */
+       p = &bp->bif_dlist;
+       while (*p != d) {
+               p = &(*p)->bd_next;
+               if (*p == 0)
+                       panic("bpf_detachd: descriptor not in list");
+       }
+       *p = (*p)->bd_next;
+       if (bp->bif_dlist == 0) {
+               /*
+                * Let the driver know that there are no more listeners.
+                */
+               /* Only call dlil_set_bpf_tap for primary dlt */
+               if (bp->bif_ifp->if_bpf == bp)
+                       dlil_set_bpf_tap(ifp, BPF_TAP_DISABLE, NULL);
+               if (bp->bif_tap)
+                       bp->bif_tap(ifp, bp->bif_dlt, BPF_TAP_DISABLE);
+               
+               for (bp = bpf_iflist; bp; bp = bp->bif_next)
+                       if (bp->bif_ifp == ifp && bp->bif_dlist != 0)
+                               break;
+               if (bp == NULL)
+                       ifp->if_bpf = NULL;
+       }
+       d->bd_bif = NULL;
        /*
         * Check if this descriptor had requested promiscuous mode.
         * If so, turn it off.
         */
        if (d->bd_promisc) {
                d->bd_promisc = 0;
-               if (ifpromisc(bp->bif_ifp, 0))
+               lck_mtx_unlock(bpf_mlock);
+               if (ifnet_set_promiscuous(ifp, 0)) {
                        /*
                         * Something is really wrong if we were able to put
                         * the driver into promiscuous mode, but can't
                         * take it out.
                         * Most likely the network interface is gone.
                         */
-                       printf("bpf: ifpromisc failed");
-       }
-       /* Remove d from the interface's descriptor list. */
-       p = &bp->bif_dlist;
-       while (*p != d) {
-               p = &(*p)->bd_next;
-               if (*p == 0)
-                       panic("bpf_detachd: descriptor not in list");
+                       printf("bpf: ifnet_set_promiscuous failed");
+               }
+               lck_mtx_lock(bpf_mlock);
        }
-       *p = (*p)->bd_next;
-       if (bp->bif_dlist == 0) {
+}
+
+
+/*
+ * Start asynchronous timer, if necessary.
+ * Must be called with bpf_mlock held.
+ */
+static void
+bpf_start_timer(struct bpf_d *d)
+{
+       uint64_t deadline;
+       struct timeval tv;
+
+       if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
+               tv.tv_sec = d->bd_rtout / hz;
+               tv.tv_usec = (d->bd_rtout % hz) * tick;
+
+               clock_interval_to_deadline((uint64_t)tv.tv_sec * USEC_PER_SEC + tv.tv_usec,
+                               NSEC_PER_USEC,
+                               &deadline);
                /*
-                * Let the driver know that there are no more listeners.
+                * The state is BPF_IDLE, so the timer hasn't 
+                * been started yet, and hasn't gone off yet;
+                * there is no thread call scheduled, so this
+                * won't change the schedule.
+                *
+                * XXX - what if, by the time it gets entered,
+                * the deadline has already passed?
                 */
-               if (ifp->if_set_bpf_tap)
-                       (*ifp->if_set_bpf_tap)(ifp, BPF_TAP_DISABLE, 0);
-               d->bd_bif->bif_ifp->if_bpf = 0;
+               thread_call_enter_delayed(d->bd_thread_call, deadline);
+               d->bd_state = BPF_WAITING;
        }
-       d->bd_bif = 0;
 }
 
+/*
+ * Cancel asynchronous timer.
+ * Must be called with bpf_mlock held.
+ */
+static boolean_t
+bpf_stop_timer(struct bpf_d *d)
+{
+       /*
+        * If the timer has already gone off, this does nothing.
+        * Our caller is expected to set d->bd_state to BPF_IDLE,
+        * with the bpf_mlock, after we are called. bpf_timed_out()
+        * also grabs bpf_mlock, so, if the timer has gone off and 
+        * bpf_timed_out() hasn't finished, it's waiting for the
+        * lock; when this thread releases the lock, it will 
+        * find the state is BPF_IDLE, and just release the 
+        * lock and return.
+        */
+       return (thread_call_cancel(d->bd_thread_call));
+}
+
+
 
 /*
  * Open ethernet device.  Returns ENXIO for illegal minor device number,
  * EBUSY if file is open by another process.
  */
 /* ARGSUSED */
-       int
-bpfopen(dev, flags, fmt, p)
-       dev_t dev;
-       int flags;
-       int fmt;
-       struct proc *p;
+int
+bpfopen(dev_t dev, int flags, __unused int fmt,
+       __unused struct proc *p)
 {
-       register struct bpf_d *d;
-
-#ifdef __APPLE__
-        /* new device nodes on demand when opening the last one */
-        if (minor(dev) == nbpfilter - 1)
-                bpf_make_dev_t(major(dev));
+       struct bpf_d *d;
 
-       if (minor(dev) >= nbpfilter)
+       lck_mtx_lock(bpf_mlock);
+       if ((unsigned int) minor(dev) >= nbpfilter) {
+               lck_mtx_unlock(bpf_mlock);
                return (ENXIO);
-       
-       d = bpf_dtab[minor(dev)];
-
-       thread_funnel_switch(KERNEL_FUNNEL, NETWORK_FUNNEL);
-#else
-       if (p->p_prison)
-               return (EPERM);
+       }
+       /* 
+        * New device nodes are created on demand when opening the last one. 
+        * The programming model is for processes to loop on the minor starting at 0 
+        * as long as EBUSY is returned. The loop stops when either the open succeeds or 
+        * an error other that EBUSY is returned. That means that bpf_make_dev_t() must 
+        * block all processes that are opening the last  node. If not all 
+        * processes are blocked, they could unexpectedly get ENOENT and abort their 
+        * opening loop.
+        */
+       if ((unsigned int) minor(dev) == (nbpfilter - 1))
+               bpf_make_dev_t(major(dev));
 
-       d = dev->si_drv1;
-#endif
        /*
         * Each minor can be opened by only one process.  If the requested 
         * minor is in use, return EBUSY.
+        *
+        * Important: bpfopen() and bpfclose() have to check and set the status of a device
+        * in the same lockin context otherwise the device may be leaked because the vnode use count 
+        * will be unpextectly greater than 1 when close() is called.
         */
-#ifdef __APPLE__
-       if (!D_ISFREE(d)) {
-            thread_funnel_switch(NETWORK_FUNNEL, KERNEL_FUNNEL);
-            return (EBUSY);
+       if (bpf_dtab[minor(dev)] == 0) {
+               bpf_dtab[minor(dev)] = (void *)1;       /* Mark opening */
+       } else {
+               lck_mtx_unlock(bpf_mlock);
+               return (EBUSY);
+       }
+       d = (struct bpf_d *)_MALLOC(sizeof(struct bpf_d), M_DEVBUF, M_WAIT);
+       if (d == NULL) {
+               /* this really is a catastrophic failure */
+               printf("bpfopen: malloc bpf_d failed\n");
+               bpf_dtab[minor(dev)] = NULL;
+               lck_mtx_unlock(bpf_mlock);
+               return ENOMEM;
        }
+       bzero(d, sizeof(struct bpf_d));
        
-       /* Mark "free" and do most initialization. */
-       bzero((char *)d, sizeof(*d));
-#else
-       if (d)
-               return (EBUSY);
-       make_dev(&bpf_cdevsw, minor(dev), 0, 0, 0600, "bpf%d", lminor(dev));
-       MALLOC(d, struct bpf_d *, sizeof(*d), M_BPF, M_WAITOK);
-       bzero(d, sizeof(*d));
-       dev->si_drv1 = d;
-#endif
+       /*
+        * It is not necessary to take the BPF lock here because no other 
+        * thread can access the device until it is marked opened...
+        */
+       
+       /* Mark "in use" and do most initialization. */
        d->bd_bufsize = bpf_bufsize;
        d->bd_sig = SIGIO;
        d->bd_seesent = 1;
-
-#ifdef __APPLE__
-       thread_funnel_switch(NETWORK_FUNNEL, KERNEL_FUNNEL);
+       d->bd_oflags = flags;
+       d->bd_state = BPF_IDLE;
+    d->bd_thread_call = thread_call_allocate(bpf_timed_out, d);
+
+       if (d->bd_thread_call == NULL) {
+               printf("bpfopen: malloc thread call failed\n");
+               bpf_dtab[minor(dev)] = NULL;
+               lck_mtx_unlock(bpf_mlock);
+               _FREE(d, M_DEVBUF);
+               return ENOMEM;
+       }
+#if CONFIG_MACF_NET
+       mac_bpfdesc_label_init(d);
+       mac_bpfdesc_label_associate(kauth_cred_get(), d);
 #endif
+       bpf_dtab[minor(dev)] = d;                               /* Mark opened */
+       lck_mtx_unlock(bpf_mlock);
 
        return (0);
 }
@@ -566,94 +715,107 @@ bpfopen(dev, flags, fmt, p)
  * deallocating its buffers, and marking it free.
  */
 /* ARGSUSED */
-       int
-bpfclose(dev, flags, fmt, p)
-       dev_t dev;
-       int flags;
-       int fmt;
-       struct proc *p;
+int
+bpfclose(dev_t dev, __unused int flags, __unused int fmt,
+        __unused struct proc *p)
 {
-       register struct bpf_d *d;
-       register int s;
-#ifdef __APPLE__
-       struct bpf_d **bpf_dtab_schk;
-#endif
+       struct bpf_d *d;
+
+       /* Take BPF lock to ensure no other thread is using the device */
+       lck_mtx_lock(bpf_mlock);
 
-#ifndef __APPLE__
-       funsetown(d->bd_sigio);
-#endif
-       s = splimp();
-#ifdef __APPLE__
-again:
        d = bpf_dtab[minor(dev)];
-       bpf_dtab_schk = bpf_dtab;
-#endif
-       thread_funnel_switch(KERNEL_FUNNEL, NETWORK_FUNNEL);
+       if (d == 0 || d == (void *)1) {
+               lck_mtx_unlock(bpf_mlock);
+               return (ENXIO);
+       }       
+       bpf_dtab[minor(dev)] = (void *)1;               /* Mark closing */
 
-#ifdef __APPLE__
        /*
-        * If someone grows bpf_dtab[] while we were waiting for the
-        * funnel, then we will be pointing off into freed memory;
-        * check to see if this is the case.
+        * Deal with any in-progress timeouts.
         */
-       if (bpf_dtab_schk != bpf_dtab) {
-               thread_funnel_switch(NETWORK_FUNNEL, KERNEL_FUNNEL);
-               goto again;
+       switch (d->bd_state) {
+               case BPF_IDLE:
+                       /*
+                        * Not waiting for a timeout, and no timeout happened.
+                        */
+                       break;
+
+               case BPF_WAITING:
+                       /*
+                        * Waiting for a timeout.
+                        * Cancel any timer that has yet to go off,
+                        * and mark the state as "closing".
+                        * Then drop the lock to allow any timers that
+                        * *have* gone off to run to completion, and wait
+                        * for them to finish.
+                        */
+                       if (!bpf_stop_timer(d)) {
+                               /*
+                                * There was no pending call, so the call must 
+                                * have been in progress. Wait for the call to
+                                * complete; we have to drop the lock while 
+                                * waiting. to let the in-progrss call complete
+                                */
+                               d->bd_state = BPF_DRAINING;
+                               while (d->bd_state == BPF_DRAINING)
+                                       msleep((caddr_t)d, bpf_mlock, PRINET,
+                                                       "bpfdraining", NULL);
+                       }
+                       d->bd_state = BPF_IDLE;
+                       break;
+
+               case BPF_TIMED_OUT:
+                       /*
+                        * Timer went off, and the timeout routine finished.
+                        */
+                       d->bd_state = BPF_IDLE;
+                       break;
+
+               case BPF_DRAINING:
+                       /*
+                        * Another thread is blocked on a close waiting for
+                        * a timeout to finish.
+                        * This "shouldn't happen", as the first thread to enter
+                        * bpfclose() will set bpf_dtab[minor(dev)] to 1, and 
+                        * all subsequent threads should see that and fail with 
+                        * ENXIO.
+                        */
+                       panic("Two threads blocked in a BPF close");
+                       break;
        }
-#endif
 
        if (d->bd_bif)
                bpf_detachd(d);
-       splx(s);
-#ifdef __APPLE__
        selthreadclear(&d->bd_sel);
+#if CONFIG_MACF_NET
+       mac_bpfdesc_label_destroy(d);
 #endif
+       thread_call_free(d->bd_thread_call);
        bpf_freed(d);
-       thread_funnel_switch(NETWORK_FUNNEL, KERNEL_FUNNEL);
+
+       /* Mark free in same context as bpfopen comes to check */
+       bpf_dtab[minor(dev)] = NULL;                    /* Mark closed */
+       lck_mtx_unlock(bpf_mlock);
+       
+       _FREE(d, M_DEVBUF);
+       
        return (0);
 }
 
-/*
- * Support for SunOS, which does not have tsleep.
- */
-#if BSD < 199103
-static
-bpf_timeout(arg)
-       caddr_t arg;
-{
-       boolean_t       funnel_state;
-       struct bpf_d *d = (struct bpf_d *)arg;
-       funnel_state = thread_funnel_set(network_flock, TRUE);
-       d->bd_timedout = 1;
-       wakeup(arg);
-       (void) thread_funnel_set(network_flock, FALSE);
-}
 
-#define BPF_SLEEP(chan, pri, s, t) bpf_sleep((struct bpf_d *)chan)
+#define BPF_SLEEP bpf_sleep
 
-int
-bpf_sleep(d)
-       register struct bpf_d *d;
+static int
+bpf_sleep(struct bpf_d *d, int pri, const char *wmesg, int timo)
 {
-       register int rto = d->bd_rtout;
-       register int st;
+       u_int64_t abstime = 0;
 
-       if (rto != 0) {
-               d->bd_timedout = 0;
-               timeout(bpf_timeout, (caddr_t)d, rto);
-       }
-       st = sleep((caddr_t)d, PRINET|PCATCH);
-       if (rto != 0) {
-               if (d->bd_timedout == 0)
-                       untimeout(bpf_timeout, (caddr_t)d);
-               else if (st == 0)
-                       return EWOULDBLOCK;
-       }
-       return (st != 0) ? EINTR : 0;
+       if(timo)
+               clock_interval_to_deadline(timo, NSEC_PER_SEC / hz, &abstime);
+       
+       return msleep1((caddr_t)d, bpf_mlock, pri, wmesg, abstime);
 }
-#else
-#define BPF_SLEEP tsleep
-#endif
 
 /*
  * Rotate the packet buffers in descriptor d.  Move the store buffer
@@ -665,43 +827,54 @@ bpf_sleep(d)
        (d)->bd_hlen = (d)->bd_slen; \
        (d)->bd_sbuf = (d)->bd_fbuf; \
        (d)->bd_slen = 0; \
-       (d)->bd_fbuf = 0;
+       (d)->bd_fbuf = NULL;
 /*
  *  bpfread - read next chunk of packets from buffers
  */
-       int
-bpfread(dev, uio, ioflag)
-       dev_t dev;
-       struct uio *uio;
-       int ioflag;
+int
+bpfread(dev_t dev, struct uio *uio, int ioflag)
 {
-       register struct bpf_d *d;
+       struct bpf_d *d;
+       int timed_out;
        int error;
-       int s;
 
-       d = bpf_dtab[minor(dev)];
+       lck_mtx_lock(bpf_mlock);
 
-       thread_funnel_switch(KERNEL_FUNNEL, NETWORK_FUNNEL);
+       d = bpf_dtab[minor(dev)];
+       if (d == 0 || d == (void *)1) {
+               lck_mtx_unlock(bpf_mlock);
+               return (ENXIO);
+       }
 
        /*
         * Restrict application to use a buffer the same size as
         * as kernel buffers.
         */
-       if (uio->uio_resid != d->bd_bufsize) {
-               thread_funnel_switch(NETWORK_FUNNEL, KERNEL_FUNNEL);
+       if (uio_resid(uio) != d->bd_bufsize) {
+               lck_mtx_unlock(bpf_mlock);
                return (EINVAL);
        }
+       
+       if (d->bd_state == BPF_WAITING)
+               bpf_stop_timer(d);
+       
+       timed_out = (d->bd_state == BPF_TIMED_OUT);
+       d->bd_state = BPF_IDLE;
 
-       s = splimp();
        /*
         * If the hold buffer is empty, then do a timed sleep, which
         * ends when the timeout expires or when enough packets
         * have arrived to fill the store buffer.
         */
        while (d->bd_hbuf == 0) {
-               if (d->bd_immediate && d->bd_slen != 0) {
+               if ((d->bd_immediate || timed_out || (ioflag & IO_NDELAY)) 
+                       && d->bd_slen != 0) {
                        /*
-                        * A packet(s) either arrived since the previous
+                        * We're in immediate mode, or are reading
+                        * in non-blocking mode, or a timer was
+                        * started before the read (e.g., by select()
+                        * or poll()) and has expired and a packet(s)
+                        * either arrived since the previous
                         * read or arrived while we were asleep.
                         * Rotate the buffers and return what's here.
                         */
@@ -716,19 +889,25 @@ bpfread(dev, uio, ioflag)
                 * it before using it again.
                 */
                if (d->bd_bif == NULL) {
-                       splx(s);
-                       thread_funnel_switch(NETWORK_FUNNEL, KERNEL_FUNNEL);
+                       lck_mtx_unlock(bpf_mlock);
+                       return (ENXIO);
+               }
+               if (ioflag & IO_NDELAY) {
+                       lck_mtx_unlock(bpf_mlock);
+                       return (EWOULDBLOCK);
+               }
+               error = BPF_SLEEP(d, PRINET|PCATCH, "bpf",
+                                 d->bd_rtout);
+               /*
+                * Make sure device is still opened
+                */
+               d = bpf_dtab[minor(dev)];
+               if (d == 0 || d == (void *)1) {
+                       lck_mtx_unlock(bpf_mlock);
                        return (ENXIO);
                }
-               
-               if (ioflag & IO_NDELAY)
-                       error = EWOULDBLOCK;
-               else
-                       error = BPF_SLEEP((caddr_t)d, PRINET|PCATCH, "bpf",
-                                         d->bd_rtout);
                if (error == EINTR || error == ERESTART) {
-                       splx(s);
-                       thread_funnel_switch(NETWORK_FUNNEL, KERNEL_FUNNEL);
+                       lck_mtx_unlock(bpf_mlock);
                        return (error);
                }
                if (error == EWOULDBLOCK) {
@@ -746,8 +925,7 @@ bpfread(dev, uio, ioflag)
                                break;
 
                        if (d->bd_slen == 0) {
-                               splx(s);
-                               thread_funnel_switch(NETWORK_FUNNEL, KERNEL_FUNNEL);
+                               lck_mtx_unlock(bpf_mlock);
                                return (0);
                        }
                        ROTATE_BUFFERS(d);
@@ -757,7 +935,6 @@ bpfread(dev, uio, ioflag)
        /*
         * At this point, we know we have something in the hold slot.
         */
-       splx(s);
 
        /*
         * Move data from hold buffer into user space.
@@ -766,12 +943,10 @@ bpfread(dev, uio, ioflag)
         */
        error = UIOMOVE(d->bd_hbuf, d->bd_hlen, UIO_READ, uio);
 
-       s = splimp();
        d->bd_fbuf = d->bd_hbuf;
-       d->bd_hbuf = 0;
+       d->bd_hbuf = NULL;
        d->bd_hlen = 0;
-       splx(s);
-       thread_funnel_switch(NETWORK_FUNNEL, KERNEL_FUNNEL);
+       lck_mtx_unlock(bpf_mlock);
        return (error);
 }
 
@@ -779,16 +954,20 @@ bpfread(dev, uio, ioflag)
 /*
  * If there are processes sleeping on this descriptor, wake them up.
  */
-static inline void
-bpf_wakeup(d)
-       register struct bpf_d *d;
+static void
+bpf_wakeup(struct bpf_d *d)
 {
+       if (d->bd_state == BPF_WAITING) {
+               bpf_stop_timer(d);
+               d->bd_state = BPF_IDLE;
+       }
        wakeup((caddr_t)d);
        if (d->bd_async && d->bd_sig && d->bd_sigio)
-               pgsigio(d->bd_sigio, d->bd_sig, 0);
+               pgsigio(d->bd_sigio, d->bd_sig);
 
 #if BSD >= 199103
        selwakeup(&d->bd_sel);
+       KNOTE(&d->bd_sel.si_note, 1);
 #ifndef __APPLE__
        /* XXX */
        d->bd_sel.si_pid = 0;
@@ -802,61 +981,130 @@ bpf_wakeup(d)
 #endif
 }
 
-/* keep in sync with bpf_movein above: */
-#define MAX_DATALINK_HDR_LEN   (sizeof(struct firewire_header))
 
-       int
-bpfwrite(dev, uio, ioflag)
-       dev_t dev;
-       struct uio *uio;
-       int ioflag;
+static void
+bpf_timed_out(void *arg, __unused void *dummy)
 {
-       register struct bpf_d *d;
-       struct ifnet *ifp;
-       struct mbuf *m;
-       int error, s;
-       char              dst_buf[SOCKADDR_HDR_LEN + MAX_DATALINK_HDR_LEN];
-       int datlen;
-
-       d = bpf_dtab[minor(dev)];
+       struct bpf_d *d = (struct bpf_d *)arg;
 
-       thread_funnel_switch(KERNEL_FUNNEL, NETWORK_FUNNEL);
-       
-       if (d->bd_bif == 0) {
-            thread_funnel_switch(NETWORK_FUNNEL, KERNEL_FUNNEL);
-            return (ENXIO);
+       lck_mtx_lock(bpf_mlock);
+       if (d->bd_state == BPF_WAITING) {
+               /*
+                * There's a select or kqueue waiting for this; if there's 
+                * now stuff to read, wake it up.
+                */
+               d->bd_state = BPF_TIMED_OUT;
+               if (d->bd_slen != 0)
+                       bpf_wakeup(d);
+       } else if (d->bd_state == BPF_DRAINING) {
+               /*
+                * A close is waiting for this to finish.
+                * Mark it as finished, and wake the close up.
+                */
+               d->bd_state = BPF_IDLE;
+               bpf_wakeup(d);
+       }
+       lck_mtx_unlock(bpf_mlock);
+}
+       
+
+
+
+
+/* keep in sync with bpf_movein above: */
+#define MAX_DATALINK_HDR_LEN   (sizeof(struct firewire_header))
+
+int
+bpfwrite(dev_t dev, struct uio *uio, __unused int ioflag)
+{
+       struct bpf_d *d;
+       struct ifnet *ifp;
+       struct mbuf *m = NULL;
+       int error;
+       char              dst_buf[SOCKADDR_HDR_LEN + MAX_DATALINK_HDR_LEN];
+       int datlen = 0;
+    int bif_dlt;
+    int bd_hdrcmplt;
+
+       lck_mtx_lock(bpf_mlock);
+
+       d = bpf_dtab[minor(dev)];
+       if (d == 0 || d == (void *)1) {
+               lck_mtx_unlock(bpf_mlock);
+               return (ENXIO);
+       }
+       if (d->bd_bif == 0) {
+               lck_mtx_unlock(bpf_mlock);
+               return (ENXIO);
        }
 
        ifp = d->bd_bif->bif_ifp;
 
-       if (uio->uio_resid == 0) {
-            thread_funnel_switch(NETWORK_FUNNEL, KERNEL_FUNNEL);
-            return (0);
+       if ((ifp->if_flags & IFF_UP) == 0) {
+               lck_mtx_unlock(bpf_mlock);
+               return (ENETDOWN);
+       }
+       if (uio_resid(uio) == 0) {
+               lck_mtx_unlock(bpf_mlock);
+               return (0);
        }
        ((struct sockaddr *)dst_buf)->sa_len = sizeof(dst_buf);
-       error = bpf_movein(uio, (int)d->bd_bif->bif_dlt, &m, 
-                          (struct sockaddr *)dst_buf, &datlen);
-       if (error) {
-            thread_funnel_switch(NETWORK_FUNNEL, KERNEL_FUNNEL);
-            return (error);
+
+   /*
+    * fix for PR-6849527
+    * geting variables onto stack before dropping lock for bpf_movein()
+    */
+    bif_dlt = (int)d->bd_bif->bif_dlt;
+    bd_hdrcmplt  = d->bd_hdrcmplt;
+       /* bpf_movein allocating mbufs; drop lock */
+    lck_mtx_unlock(bpf_mlock);
+
+       error = bpf_movein(uio, bif_dlt, &m, 
+    bd_hdrcmplt ? NULL : (struct sockaddr *)dst_buf,
+    &datlen);
+       
+    if (error) {
+               return (error);
        }
 
-       if (datlen > ifp->if_mtu) {
-            thread_funnel_switch(NETWORK_FUNNEL, KERNEL_FUNNEL);
-            return (EMSGSIZE);
+       /* taking the lock again and verifying whether device is open */
+    lck_mtx_lock(bpf_mlock);
+       d = bpf_dtab[minor(dev)];
+       if (d == 0 || d == (void *)1) {
+               lck_mtx_unlock(bpf_mlock);
+               m_freem(m);
+               return (ENXIO);
        }
 
-       if (d->bd_hdrcmplt) {
-               ((struct sockaddr *)dst_buf)->sa_family = pseudo_AF_HDRCMPLT;
+       if (d->bd_bif == NULL) {
+               lck_mtx_unlock(bpf_mlock);
+               m_free(m);
+               return (ENXIO);
+       }
+
+       if ((unsigned)datlen > ifp->if_mtu) {
+               lck_mtx_unlock(bpf_mlock);
+               m_freem(m);
+               return (EMSGSIZE);
        }
 
-       s = splnet();
 
-       error = dlil_output(ifptodlt(ifp, PF_INET), m, 
-                       (caddr_t) 0, (struct sockaddr *)dst_buf, 0);
+#if CONFIG_MACF_NET
+       mac_mbuf_label_associate_bpfdesc(d, m);
+#endif
+       lck_mtx_unlock(bpf_mlock);
+
+       if (d->bd_hdrcmplt) {
+               if (d->bd_bif->bif_send)
+                       error = d->bd_bif->bif_send(ifp, d->bd_bif->bif_dlt, m);
+               else
+                       error = dlil_output(ifp, 0, m, NULL, NULL, 1);
+       }
+       else {
+               error = dlil_output(ifp, PF_INET, m, NULL, (struct sockaddr *)dst_buf, 0);
+       }
 
-       splx(s);
-       thread_funnel_switch(NETWORK_FUNNEL, KERNEL_FUNNEL);
        /*
         * The driver frees the mbuf.
         */
@@ -865,16 +1113,15 @@ bpfwrite(dev, uio, ioflag)
 
 /*
  * Reset a descriptor by flushing its packet buffer and clearing the
- * receive and drop counts.  Should be called at splimp.
+ * receive and drop counts.
  */
 static void
-reset_d(d)
-       struct bpf_d *d;
+reset_d(struct bpf_d *d)
 {
        if (d->bd_hbuf) {
                /* Free the hold buffer. */
                d->bd_fbuf = d->bd_hbuf;
-               d->bd_hbuf = 0;
+               d->bd_hbuf = NULL;
        }
        d->bd_slen = 0;
        d->bd_hlen = 0;
@@ -904,19 +1151,23 @@ reset_d(d)
  */
 /* ARGSUSED */
 int
-bpfioctl(dev, cmd, addr, flags, p)
-       dev_t dev;
-       u_long cmd;
-       caddr_t addr;
-       int flags;
-       struct proc *p;
+bpfioctl(dev_t dev, u_long cmd, caddr_t addr, __unused int flags,
+    struct proc *p)
 {
-       register struct bpf_d *d;
-       int s, error = 0;
+       struct bpf_d *d;
+       int error = 0;
+
+       lck_mtx_lock(bpf_mlock);
 
        d = bpf_dtab[minor(dev)];
+       if (d == 0 || d == (void *)1) {
+               lck_mtx_unlock(bpf_mlock);
+               return (ENXIO);
+       }
 
-       thread_funnel_switch(KERNEL_FUNNEL, NETWORK_FUNNEL);
+       if (d->bd_state == BPF_WAITING)
+               bpf_stop_timer(d);
+       d->bd_state = BPF_IDLE;
 
        switch (cmd) {
 
@@ -931,11 +1182,9 @@ bpfioctl(dev, cmd, addr, flags, p)
                {
                        int n;
 
-                       s = splimp();
                        n = d->bd_slen;
                        if (d->bd_hbuf)
                                n += d->bd_hlen;
-                       splx(s);
 
                        *(int *)addr = n;
                        break;
@@ -949,7 +1198,7 @@ bpfioctl(dev, cmd, addr, flags, p)
                                error = EINVAL;
                        else {
                                ifp = d->bd_bif->bif_ifp;
-                               error = (*ifp->if_ioctl)(ifp, cmd, addr);
+                               error = ifnet_ioctl(ifp, 0, cmd, addr);
                        }
                        break;
                }
@@ -971,7 +1220,7 @@ bpfioctl(dev, cmd, addr, flags, p)
                if (d->bd_bif != 0)
                        error = EINVAL;
                else {
-                       register u_int size = *(u_int *)addr;
+                       u_int size = *(u_int *)addr;
 
                        if (size > bpf_maxbufsize)
                                *(u_int *)addr = size = bpf_maxbufsize;
@@ -985,17 +1234,24 @@ bpfioctl(dev, cmd, addr, flags, p)
        /*
         * Set link layer read filter.
         */
-       case BIOCSETF:
-               error = bpf_setf(d, (struct bpf_program *)addr);
+       case BIOCSETF32: {
+               struct bpf_program32 *prg32 = (struct bpf_program32 *)addr;
+               error = bpf_setf(d, prg32->bf_len,
+                   CAST_USER_ADDR_T(prg32->bf_insns));
+               break;
+       }
+
+       case BIOCSETF64: {
+               struct bpf_program64 *prg64 = (struct bpf_program64 *)addr;
+               error = bpf_setf(d, prg64->bf_len, prg64->bf_insns);
                break;
+       }
 
        /*
         * Flush read packet buffer.
         */
        case BIOCFLUSH:
-               s = splimp();
                reset_d(d);
-               splx(s);
                break;
 
        /*
@@ -1009,13 +1265,13 @@ bpfioctl(dev, cmd, addr, flags, p)
                        error = EINVAL;
                        break;
                }
-               s = splimp();
                if (d->bd_promisc == 0) {
-                       error = ifpromisc(d->bd_bif->bif_ifp, 1);
+                       lck_mtx_unlock(bpf_mlock);
+                       error = ifnet_set_promiscuous(d->bd_bif->bif_ifp, 1);
+                       lck_mtx_lock(bpf_mlock);
                        if (error == 0)
                                d->bd_promisc = 1;
                }
-               splx(s);
                break;
 
        /*
@@ -1028,6 +1284,28 @@ bpfioctl(dev, cmd, addr, flags, p)
                        *(u_int *)addr = d->bd_bif->bif_dlt;
                break;
 
+       /*
+        * Get a list of supported data link types.
+        */
+       case BIOCGDLTLIST:
+               if (d->bd_bif == NULL) {
+                       error = EINVAL;
+               } else {
+                       error = bpf_getdltlist(d,
+                           (struct bpf_dltlist *)addr, p);
+               }
+               break;
+
+       /*
+        * Set data link type.
+        */
+       case BIOCSDLT:
+                       if (d->bd_bif == NULL)
+                                       error = EINVAL;
+                       else
+                                       error = bpf_setdlt(d, *(u_int *)addr);
+                       break;
+
        /*
         * Get interface name.
         */
@@ -1046,37 +1324,73 @@ bpfioctl(dev, cmd, addr, flags, p)
        /*
         * Set interface.
         */
-       case BIOCSETIF:
-               error = bpf_setif(d, (struct ifreq *)addr);
+       case BIOCSETIF: {
+               ifnet_t ifp;
+               ifp = ifunit(((struct ifreq *)addr)->ifr_name);
+               if (ifp == NULL)
+                       error = ENXIO;
+               else
+                       error = bpf_setif(d, ifp, 0);
                break;
+       }
 
        /*
         * Set read timeout.
         */
-       case BIOCSRTIMEOUT:
-               {
-                       struct timeval *tv = (struct timeval *)addr;
+        case BIOCSRTIMEOUT32:
+                {
+                       struct user32_timeval *_tv = (struct user32_timeval *)addr;
+                       struct timeval tv;
+
+                       tv.tv_sec  = _tv->tv_sec;
+                       tv.tv_usec = _tv->tv_usec;
 
+                        /*
+                        * Subtract 1 tick from tvtohz() since this isn't
+                        * a one-shot timer.
+                        */
+                       if ((error = itimerfix(&tv)) == 0)
+                               d->bd_rtout = tvtohz(&tv) - 1;
+                       break;
+                }
+
+        case BIOCSRTIMEOUT64:
+                {
+                       struct user64_timeval *_tv = (struct user64_timeval *)addr;
+                       struct timeval tv;
+                        
+                       tv.tv_sec  = _tv->tv_sec;
+                       tv.tv_usec = _tv->tv_usec;
+                        
                        /*
                         * Subtract 1 tick from tvtohz() since this isn't
                         * a one-shot timer.
                         */
-                       if ((error = itimerfix(tv)) == 0)
-                               d->bd_rtout = tvtohz(tv) - 1;
+                       if ((error = itimerfix(&tv)) == 0)
+                               d->bd_rtout = tvtohz(&tv) - 1;
                        break;
-               }
-
-       /*
+                }
+       
+        /*
         * Get read timeout.
         */
-       case BIOCGRTIMEOUT:
+       case BIOCGRTIMEOUT32:
                {
-                       struct timeval *tv = (struct timeval *)addr;
+                       struct user32_timeval *tv = (struct user32_timeval *)addr;
 
                        tv->tv_sec = d->bd_rtout / hz;
                        tv->tv_usec = (d->bd_rtout % hz) * tick;
                        break;
-               }
+                }
+
+       case BIOCGRTIMEOUT64:
+               {
+                       struct user64_timeval *tv = (struct user64_timeval *)addr;
+
+                       tv->tv_sec = d->bd_rtout / hz;
+                       tv->tv_usec = (d->bd_rtout % hz) * tick;
+                       break;
+                }
 
        /*
         * Get packet stats.
@@ -1175,7 +1489,9 @@ bpfioctl(dev, cmd, addr, flags, p)
                *(u_int *)addr = d->bd_sig;
                break;
        }
-       thread_funnel_switch(NETWORK_FUNNEL, KERNEL_FUNNEL);
+       
+       lck_mtx_unlock(bpf_mlock);
+
        return (error);
 }
 
@@ -1184,42 +1500,35 @@ bpfioctl(dev, cmd, addr, flags, p)
  * free it and replace it.  Returns EINVAL for bogus requests.
  */
 static int
-bpf_setf(d, fp)
-       struct bpf_d *d;
-       struct bpf_program *fp;
+bpf_setf(struct bpf_d *d, u_int bf_len, user_addr_t bf_insns)
 {
        struct bpf_insn *fcode, *old;
        u_int flen, size;
-       int s;
 
        old = d->bd_filter;
-       if (fp->bf_insns == 0) {
-               if (fp->bf_len != 0)
+       if (bf_insns == USER_ADDR_NULL) {
+               if (bf_len != 0)
                        return (EINVAL);
-               s = splimp();
-               d->bd_filter = 0;
+               d->bd_filter = NULL;
                reset_d(d);
-               splx(s);
                if (old != 0)
                        FREE((caddr_t)old, M_DEVBUF);
                return (0);
        }
-       flen = fp->bf_len;
+       flen = bf_len;
        if (flen > BPF_MAXINSNS)
                return (EINVAL);
 
-       size = flen * sizeof(*fp->bf_insns);
+       size = flen * sizeof(struct bpf_insn);
        fcode = (struct bpf_insn *) _MALLOC(size, M_DEVBUF, M_WAIT);
 #ifdef __APPLE__
        if (fcode == NULL)
                return (ENOBUFS);
 #endif
-       if (copyin((caddr_t)fp->bf_insns, (caddr_t)fcode, size) == 0 &&
+       if (copyin(bf_insns, (caddr_t)fcode, size) == 0 &&
            bpf_validate(fcode, (int)flen)) {
-               s = splimp();
                d->bd_filter = fcode;
                reset_d(d);
-               splx(s);
                if (old != 0)
                        FREE((caddr_t)old, M_DEVBUF);
 
@@ -1235,42 +1544,30 @@ bpf_setf(d, fp)
  * Return an errno or 0.
  */
 static int
-bpf_setif(d, ifr)
-       struct bpf_d *d;
-       struct ifreq *ifr;
+bpf_setif(struct bpf_d *d, ifnet_t theywant, u_int32_t dlt)
 {
        struct bpf_if *bp;
-       int s, error;
-       struct ifnet *theywant;
-
-       theywant = ifunit(ifr->ifr_name);
-       if (theywant == 0)
-               return ENXIO;
-
+       int error;
+       
        /*
         * Look through attached interfaces for the named one.
         */
        for (bp = bpf_iflist; bp != 0; bp = bp->bif_next) {
                struct ifnet *ifp = bp->bif_ifp;
 
-               if (ifp == 0 || ifp != theywant)
+               if (ifp == 0 || ifp != theywant || (dlt != 0 && dlt != bp->bif_dlt))
                        continue;
                /*
                 * We found the requested interface.
-                * If it's not up, return an error.
                 * Allocate the packet buffers if we need to.
                 * If we're already attached to requested interface,
                 * just flush the buffer.
                 */
-               if ((ifp->if_flags & IFF_UP) == 0)
-                       return (ENETDOWN);
-
                if (d->bd_sbuf == 0) {
                        error = bpf_allocbufs(d);
                        if (error != 0)
                                return (error);
                }
-               s = splimp();
                if (bp != d->bd_bif) {
                        if (d->bd_bif)
                                /*
@@ -1278,91 +1575,283 @@ bpf_setif(d, ifr)
                                 */
                                bpf_detachd(d);
 
-                       bpf_attachd(d, bp);
+                       if (bpf_attachd(d, bp) != 0) {
+                               return ENXIO;
+                       }
                }
                reset_d(d);
-               splx(s);
                return (0);
        }
        /* Not found. */
        return (ENXIO);
 }
 
+
+
+/*
+ * Get a list of available data link type of the interface.
+ */
+static int
+bpf_getdltlist(struct bpf_d *d, struct bpf_dltlist *bfl, struct proc *p)
+{
+       u_int           n;
+       int             error;
+       struct ifnet    *ifp;
+       struct bpf_if   *bp;
+       user_addr_t     dlist;
+
+       if (proc_is64bit(p)) {
+               dlist = (user_addr_t)bfl->bfl_u.bflu_pad;
+       } else {
+               dlist = CAST_USER_ADDR_T(bfl->bfl_u.bflu_list);
+       }
+
+       ifp = d->bd_bif->bif_ifp;
+       n = 0;
+       error = 0;
+       for (bp = bpf_iflist; bp; bp = bp->bif_next) {
+               if (bp->bif_ifp != ifp)
+                       continue;
+               if (dlist != USER_ADDR_NULL) {
+                       if (n >= bfl->bfl_len) {
+                               return (ENOMEM);
+                       }
+                       error = copyout(&bp->bif_dlt, dlist,
+                           sizeof (bp->bif_dlt));
+                       dlist += sizeof (bp->bif_dlt);
+               }
+               n++;
+       }
+       bfl->bfl_len = n;
+       return (error);
+}
+
 /*
- * Support for select() and poll() system calls
+ * Set the data link type of a BPF instance.
+ */
+static int
+bpf_setdlt(struct bpf_d *d, uint32_t dlt)
+       
+
+{
+       int error, opromisc;
+       struct ifnet *ifp;
+       struct bpf_if *bp;
+       
+       if (d->bd_bif->bif_dlt == dlt)
+               return (0);
+       ifp = d->bd_bif->bif_ifp;
+       for (bp = bpf_iflist; bp; bp = bp->bif_next) {
+               if (bp->bif_ifp == ifp && bp->bif_dlt == dlt)
+                       break;
+       }
+       if (bp != NULL) {
+               opromisc = d->bd_promisc;
+               bpf_detachd(d);
+               error = bpf_attachd(d, bp);
+               if (error) {
+                       printf("bpf_setdlt: bpf_attachd %s%d failed (%d)\n",
+                               ifnet_name(bp->bif_ifp), ifnet_unit(bp->bif_ifp), error);
+                       return error;
+               }
+               reset_d(d);
+               if (opromisc) {
+                       lck_mtx_unlock(bpf_mlock);
+                       error = ifnet_set_promiscuous(bp->bif_ifp, 1);
+                       lck_mtx_lock(bpf_mlock);
+                       if (error)
+                               printf("bpf_setdlt: ifpromisc %s%d failed (%d)\n",
+                                          ifnet_name(bp->bif_ifp), ifnet_unit(bp->bif_ifp), error);
+                       else
+                               d->bd_promisc = 1;
+               }
+       }
+       return (bp == NULL ? EINVAL : 0);
+}
+
+/*
+ * Support for select()
  *
  * Return true iff the specific operation will not block indefinitely.
  * Otherwise, return false but make a note that a selwakeup() must be done.
  */
 int
-bpfpoll(dev, events, wql, p)
-       register dev_t dev;
-       int events;
-       void * wql;
-       struct proc *p;
+bpfselect(dev_t dev, int which, void * wql, struct proc *p)
 {
-       register struct bpf_d *d;
-       register int s;
-       int revents = 0;
+       struct bpf_d *d;
+       int ret = 0;
+
+       lck_mtx_lock(bpf_mlock);
 
        d = bpf_dtab[minor(dev)];
+       if (d == 0 || d == (void *)1) {
+               lck_mtx_unlock(bpf_mlock);
+               return (ENXIO);
+       }
 
-       thread_funnel_switch(KERNEL_FUNNEL, NETWORK_FUNNEL);
-       /*
-        * An imitation of the FIONREAD ioctl code.
-        */
        if (d->bd_bif == NULL) {
-               thread_funnel_switch(NETWORK_FUNNEL, KERNEL_FUNNEL);
+               lck_mtx_unlock(bpf_mlock);
                return (ENXIO);
        }
 
-       s = splimp();
-       if (events & (POLLIN | POLLRDNORM)) {
-               if (d->bd_hlen != 0 || (d->bd_immediate && d->bd_slen != 0))
-                       revents |= events & (POLLIN | POLLRDNORM);
-               else
-                       selrecord(p, &d->bd_sel, wql);
+       switch (which) {
+               case FREAD:
+                       if (d->bd_hlen != 0 ||
+                                       ((d->bd_immediate || d->bd_state == BPF_TIMED_OUT) &&
+                                        d->bd_slen != 0))
+                               ret = 1; /* read has data to return */
+                       else {
+                               /*
+                                * Read has no data to return.
+                                * Make the select wait, and start a timer if
+                                * necessary.
+                                */
+                               selrecord(p, &d->bd_sel, wql);
+                               bpf_start_timer(d);
+                       }
+                       break;
+
+               case FWRITE:
+                       ret = 1; /* can't determine whether a write would block */
+                       break;
        }
-       splx(s);
-       thread_funnel_switch(NETWORK_FUNNEL, KERNEL_FUNNEL);
-       return (revents);
+
+       lck_mtx_unlock(bpf_mlock);
+       return (ret);
 }
 
+
 /*
- * Incoming linkage from device drivers.  Process the packet pkt, of length
- * pktlen, which is stored in a contiguous buffer.  The packet is parsed
- * by each process' filter, and if accepted, stashed into the corresponding
- * buffer.
+ * Support for kevent() system call.  Register EVFILT_READ filters and
+ * reject all others.
  */
-void
-bpf_tap(ifp, pkt, pktlen)
-       struct ifnet *ifp;
-       register u_char *pkt;
-       register u_int pktlen;
+int bpfkqfilter(dev_t dev, struct knote *kn);
+static void filt_bpfdetach(struct knote *);
+static int filt_bpfread(struct knote *, long);
+
+static struct filterops bpfread_filtops = {
+       .f_isfd = 1, 
+       .f_detach = filt_bpfdetach,
+       .f_event = filt_bpfread,
+};
+
+int
+bpfkqfilter(dev_t dev, struct knote *kn)
 {
-       struct bpf_if *bp;
-       register struct bpf_d *d;
-       register u_int slen;
+       struct bpf_d *d;
+
        /*
-        * Note that the ipl does not have to be raised at this point.
-        * The only problem that could arise here is that if two different
-        * interfaces shared any data.  This is not the case.
+        * Is this device a bpf?
         */
-       thread_funnel_switch(KERNEL_FUNNEL, NETWORK_FUNNEL);
-       bp = ifp->if_bpf;
-#ifdef __APPLE__
-       if (bp) {
-#endif
-       for (d = bp->bif_dlist; d != 0; d = d->bd_next) {
-               ++d->bd_rcount;
-               slen = bpf_filter(d->bd_filter, pkt, pktlen, pktlen);
-               if (slen != 0)
-                       catchpacket(d, pkt, pktlen, slen, bcopy);
+       if (major(dev) != CDEV_MAJOR) {
+               return (EINVAL);
        }
-#ifdef __APPLE__
-    }
-       thread_funnel_switch(KERNEL_FUNNEL, NETWORK_FUNNEL);
-#endif
+
+       if (kn->kn_filter != EVFILT_READ) {
+               return (EINVAL);
+       }
+
+       lck_mtx_lock(bpf_mlock);
+
+       d = bpf_dtab[minor(dev)];
+       if (d == 0 || d == (void *)1) {
+               lck_mtx_unlock(bpf_mlock);
+               return (ENXIO);
+       }
+
+       if (d->bd_bif == NULL) {
+               lck_mtx_unlock(bpf_mlock);
+               return (ENXIO);
+       }
+
+       kn->kn_hook = d;
+       kn->kn_fop = &bpfread_filtops;
+       KNOTE_ATTACH(&d->bd_sel.si_note, kn);
+       lck_mtx_unlock(bpf_mlock);
+       return 0;
+}
+
+static void
+filt_bpfdetach(struct knote *kn)
+{
+       struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
+
+       lck_mtx_lock(bpf_mlock);
+       KNOTE_DETACH(&d->bd_sel.si_note, kn);
+       lck_mtx_unlock(bpf_mlock);
+}
+
+static int
+filt_bpfread(struct knote *kn, long hint)
+{
+       struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
+       int ready = 0;
+
+       if (hint == 0)
+               lck_mtx_lock(bpf_mlock);
+
+       if (d->bd_immediate) {
+               /*
+                * If there's data in the hold buffer, it's the 
+                * amount of data a read will return.
+                *
+                * If there's no data in the hold buffer, but
+                * there's data in the store buffer, a read will
+                * immediately rotate the store buffer to the 
+                * hold buffer, the amount of data in the store
+                * buffer is the amount of data a read will 
+                * return.
+                *
+                * If there's no data in either buffer, we're not 
+                * ready to read.
+                */
+               kn->kn_data = (d->bd_hlen == 0 ? d->bd_slen : d->bd_hlen);
+               int64_t lowwat = 1;
+               if (kn->kn_sfflags & NOTE_LOWAT)
+               {
+                       if (kn->kn_sdata > d->bd_bufsize)
+                               lowwat = d->bd_bufsize;
+                       else if (kn->kn_sdata > lowwat)
+                               lowwat = kn->kn_sdata;
+               }
+               ready = (kn->kn_data >= lowwat);
+       } else {
+               /*
+                * If there's data in the hold buffer, it's the 
+                * amount of data a read will return.
+                *
+                * If there's no data in the hold buffer, but 
+                * there's data in the store buffer, if the 
+                * timer has expired a read will immediately
+                * rotate the store buffer to the hold buffer,
+                * so the amount of data in the store buffer is 
+                * the amount of data a read will return.
+                *
+                * If there's no data in either buffer, or there's 
+                * no data in the hold buffer and the timer hasn't 
+                * expired, we're not ready to read.
+                */
+               kn->kn_data = (d->bd_hlen == 0 && d->bd_state == BPF_TIMED_OUT ? 
+                               d->bd_slen : d->bd_hlen);
+               ready = (kn->kn_data > 0);
+       }
+       if (!ready)
+               bpf_start_timer(d);
+
+       if (hint == 0)
+               lck_mtx_unlock(bpf_mlock);
+       return (ready);
+}
+
+static inline void*
+_cast_non_const(const void * ptr) {
+       union {
+               const void*             cval;
+               void*                   val;
+       } ret;
+       
+       ret.cval = ptr;
+       return (ret.val);
 }
 
 /*
@@ -1370,53 +1859,135 @@ bpf_tap(ifp, pkt, pktlen)
  * from m_copydata in sys/uipc_mbuf.c.
  */
 static void
-bpf_mcopy(src_arg, dst_arg, len)
-       const void *src_arg;
-       void *dst_arg;
-       register size_t len;
+bpf_mcopy(const void *src_arg, void *dst_arg, size_t len)
 {
-       register const struct mbuf *m;
-       register u_int count;
+       struct mbuf *m = _cast_non_const(src_arg);
+       u_int count;
        u_char *dst;
 
-       m = src_arg;
        dst = dst_arg;
        while (len > 0) {
                if (m == 0)
                        panic("bpf_mcopy");
                count = min(m->m_len, len);
-               bcopy(mtod((struct mbuf *)m, void *), dst, count);
+               bcopy(mbuf_data(m), dst, count);
                m = m->m_next;
                dst += count;
                len -= count;
        }
 }
 
-/*
- * Incoming linkage from device drivers, when packet is in an mbuf chain.
- */
-void
-bpf_mtap(ifp, m)
-       struct ifnet *ifp;
-       struct mbuf *m;
+static inline void
+bpf_tap_imp(
+       ifnet_t         ifp,
+       u_int32_t       dlt,
+       mbuf_t          m,
+       void*           hdr,
+       size_t          hlen,
+       int                     outbound)
 {
-       struct bpf_if *bp = ifp->if_bpf;
-       struct bpf_d *d;
-       u_int pktlen, slen;
-       struct mbuf *m0;
+       struct bpf_if *bp;
 
-       pktlen = 0;
-       for (m0 = m; m0 != 0; m0 = m0->m_next)
-               pktlen += m0->m_len;
+       /*
+        * It's possible that we get here after the bpf descriptor has been
+        * detached from the interface; in such a case we simply return.
+        * Lock ordering is important since we can be called asynchronously
+        * (from the IOKit) to process an inbound packet; when that happens
+        * we would have been holding its "gateLock" and will be acquiring
+        * "bpf_mlock" upon entering this routine.  Due to that, we release
+        * "bpf_mlock" prior to calling ifnet_set_promiscuous (which will
+        * acquire "gateLock" in the IOKit), in order to avoid a deadlock
+        * when a ifnet_set_promiscuous request simultaneously collides with
+        * an inbound packet being passed into the tap callback.
+        */
+       lck_mtx_lock(bpf_mlock);
+       if (ifp->if_bpf == NULL) {
+               lck_mtx_unlock(bpf_mlock);
+               return;
+       }
+       bp = ifp->if_bpf;
+       for (bp = ifp->if_bpf; bp && bp->bif_ifp == ifp &&
+                (dlt != 0 && bp->bif_dlt != dlt); bp = bp->bif_next)
+               ;
+       if (bp && bp->bif_ifp == ifp && bp->bif_dlist != NULL) {
+               struct bpf_d    *d;
+               struct m_hdr    hack_hdr;
+               u_int   pktlen = 0;
+               u_int   slen = 0;
+               struct mbuf *m0;
+               
+               if (hdr) {
+                       /*
+                        * This is gross. We mock up an mbuf that points to the
+                        * header buffer. This means we don't have to copy the
+                        * header. A number of interfaces prepended headers just
+                        * for bpf by allocating an mbuf on the stack. We want to
+                        * give developers an easy way to prepend a header for bpf.
+                        * Since a developer allocating an mbuf on the stack is bad,
+                        * we do even worse here, allocating only a header to point
+                        * to a buffer the developer supplied. This makes assumptions
+                        * that bpf_filter and catchpacket will not look at anything
+                        * in the mbuf other than the header. This was true at the
+                        * time this code was written.
+                        */
+                       hack_hdr.mh_next = m;
+                       hack_hdr.mh_nextpkt = NULL;
+                       hack_hdr.mh_len = hlen;
+                       hack_hdr.mh_data = hdr;
+                       hack_hdr.mh_type = m->m_type;
+                       hack_hdr.mh_flags = 0;
+                       
+                       m = (mbuf_t)&hack_hdr;
+               }
 
-       for (d = bp->bif_dlist; d != 0; d = d->bd_next) {
-               if (!d->bd_seesent && (m->m_pkthdr.rcvif == NULL))
-                       continue;
-               ++d->bd_rcount;
-               slen = bpf_filter(d->bd_filter, (u_char *)m, pktlen, 0);
-               if (slen != 0)
-                       catchpacket(d, (u_char *)m, pktlen, slen, bpf_mcopy);
+               for (m0 = m; m0 != 0; m0 = m0->m_next)
+                       pktlen += m0->m_len;
+               
+               for (d = bp->bif_dlist; d; d = d->bd_next) {
+                       if (outbound && !d->bd_seesent)
+                               continue;
+                       ++d->bd_rcount;
+                       slen = bpf_filter(d->bd_filter, (u_char *)m, pktlen, 0);
+                       if (slen != 0) {
+#if CONFIG_MACF_NET
+                               if (mac_bpfdesc_check_receive(d, bp->bif_ifp) != 0)
+                                       continue;
+#endif
+                               catchpacket(d, (u_char *)m, pktlen, slen, bpf_mcopy);
+                       }
+               }
        }
+       lck_mtx_unlock(bpf_mlock);
+}
+
+void
+bpf_tap_out(
+       ifnet_t         ifp,
+       u_int32_t       dlt,
+       mbuf_t          m,
+       void*           hdr,
+       size_t          hlen)
+{
+       bpf_tap_imp(ifp, dlt, m, hdr, hlen, 1);
+}
+
+void
+bpf_tap_in(
+       ifnet_t         ifp,
+       u_int32_t       dlt,
+       mbuf_t          m,
+       void*           hdr,
+       size_t          hlen)
+{
+       bpf_tap_imp(ifp, dlt, m, hdr, hlen, 0);
+}
+
+/* Callback registered with Ethernet driver. */
+static int bpf_tap_callback(struct ifnet *ifp, struct mbuf *m)
+{
+       bpf_tap_imp(ifp, 0, m, NULL, 0, mbuf_pkthdr_rcvif(m) == NULL);
+       
+       return 0;
 }
 
 /*
@@ -1428,15 +1999,13 @@ bpf_mtap(ifp, m)
  * pkt is really an mbuf.
  */
 static void
-catchpacket(d, pkt, pktlen, snaplen, cpfn)
-       register struct bpf_d *d;
-       register u_char *pkt;
-       register u_int pktlen, snaplen;
-       register void (*cpfn) __P((const void *, void *, size_t));
+catchpacket(struct bpf_d *d, u_char *pkt, u_int pktlen, u_int snaplen, 
+       void (*cpfn)(const void *, void *, size_t))
 {
-       register struct bpf_hdr *hp;
-       register int totlen, curlen;
-       register int hdrlen = d->bd_bif->bif_hdrlen;
+       struct bpf_hdr *hp;
+       int totlen, curlen;
+       int hdrlen = d->bd_bif->bif_hdrlen;
+       int do_wakeup = 0;
        /*
         * Figure out how many bytes to move.  If the packet is
         * greater or equal to the snapshot length, transfer that
@@ -1457,7 +2026,7 @@ catchpacket(d, pkt, pktlen, snaplen, cpfn)
                 * Rotate the buffers if we can, then wakeup any
                 * pending reads.
                 */
-               if (d->bd_fbuf == 0) {
+               if (d->bd_fbuf == NULL) {
                        /*
                         * We haven't completed the previous read yet,
                         * so drop the packet.
@@ -1466,27 +2035,25 @@ catchpacket(d, pkt, pktlen, snaplen, cpfn)
                        return;
                }
                ROTATE_BUFFERS(d);
-               bpf_wakeup(d);
+               do_wakeup = 1;
                curlen = 0;
        }
-       else if (d->bd_immediate)
+       else if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT)
                /*
-                * Immediate mode is set.  A packet arrived so any
-                * reads should be woken up.
+                * Immediate mode is set, or the read timeout has 
+                * already expired during a select call. A packet 
+                * arrived, so the reader should be woken up.
                 */
-               bpf_wakeup(d);
+               do_wakeup = 1;
 
        /*
         * Append the bpf header.
         */
        hp = (struct bpf_hdr *)(d->bd_sbuf + curlen);
-#if BSD >= 199103
-       microtime(&hp->bh_tstamp);
-#elif defined(sun)
-       uniqtime(&hp->bh_tstamp);
-#else
-       hp->bh_tstamp = time;
-#endif
+       struct timeval tv;
+       microtime(&tv);
+       hp->bh_tstamp.tv_sec = tv.tv_sec;
+       hp->bh_tstamp.tv_usec = tv.tv_usec;
        hp->bh_datalen = pktlen;
        hp->bh_hdrlen = hdrlen;
        /*
@@ -1494,14 +2061,16 @@ catchpacket(d, pkt, pktlen, snaplen, cpfn)
         */
        (*cpfn)(pkt, (u_char *)hp + hdrlen, (hp->bh_caplen = totlen - hdrlen));
        d->bd_slen = curlen + totlen;
+
+       if (do_wakeup)
+               bpf_wakeup(d);
 }
 
 /*
  * Initialize all nonzero fields of a descriptor.
  */
 static int
-bpf_allocbufs(d)
-       register struct bpf_d *d;
+bpf_allocbufs(struct bpf_d *d)
 {
        d->bd_fbuf = (caddr_t) _MALLOC(d->bd_bufsize, M_DEVBUF, M_WAIT);
        if (d->bd_fbuf == 0)
@@ -1522,8 +2091,7 @@ bpf_allocbufs(d)
  * Called on close.
  */
 static void
-bpf_freed(d)
-       register struct bpf_d *d;
+bpf_freed(struct bpf_d *d)
 {
        /*
         * We don't need to lock out interrupts since this descriptor has
@@ -1539,8 +2107,6 @@ bpf_freed(d)
        }
        if (d->bd_filter)
                FREE((caddr_t)d->bd_filter, M_DEVBUF);
-
-       D_MARKFREE(d);
 }
 
 /*
@@ -1549,37 +2115,83 @@ bpf_freed(d)
  * size of the link header (variable length headers not yet supported).
  */
 void
-bpfattach(ifp, dlt, hdrlen)
-       struct ifnet *ifp;
-       u_int dlt, hdrlen;
+bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
 {
-       struct bpf_if *bp;
-       int i;
-       bp = (struct bpf_if *) _MALLOC(sizeof(*bp), M_DEVBUF, M_WAIT);
-       if (bp == 0)
-               panic("bpfattach");
-
-       bp->bif_dlist = 0;
-       bp->bif_ifp = ifp;
-       bp->bif_dlt = dlt;
+       bpf_attach(ifp, dlt, hdrlen, NULL, NULL);
+}
 
-       bp->bif_next = bpf_iflist;
-       bpf_iflist = bp;
+errno_t
+bpf_attach(
+       ifnet_t                 ifp,
+       u_int32_t               dlt,
+       u_int32_t               hdrlen,
+       bpf_send_func   send,
+       bpf_tap_func    tap)
+{
+       struct bpf_if *bp_new;
+       struct bpf_if *bp_temp;
+       struct bpf_if *bp_first = NULL;
+       
+       bp_new = (struct bpf_if *) _MALLOC(sizeof(*bp_new), M_DEVBUF, M_WAIT);
+       if (bp_new == 0)
+               panic("bpfattach");
 
-       bp->bif_ifp->if_bpf = 0;
+       lck_mtx_lock(bpf_mlock);
 
+       /*
+        * Check if this interface/dlt is already attached, record first
+        * attachment for this interface.
+        */
+       for (bp_temp = bpf_iflist; bp_temp && (bp_temp->bif_ifp != ifp ||
+                bp_temp->bif_dlt != dlt); bp_temp = bp_temp->bif_next) {
+                if (bp_temp->bif_ifp == ifp && bp_first == NULL)
+                       bp_first = bp_temp;
+       }
+       
+       if (bp_temp != NULL) {
+               printf("bpfattach - %s%d with dlt %d is already attached\n",
+                       ifp->if_name, ifp->if_unit, dlt);
+               FREE(bp_new, M_DEVBUF);
+               lck_mtx_unlock(bpf_mlock);
+               return EEXIST;
+       }
+       
+       bzero(bp_new, sizeof(*bp_new));
+       bp_new->bif_ifp = ifp;
+       bp_new->bif_dlt = dlt;
+       bp_new->bif_send = send;
+       bp_new->bif_tap = tap;
+       
+       if (bp_first == NULL) {
+               /* No other entries for this ifp */
+               bp_new->bif_next = bpf_iflist;
+               bpf_iflist = bp_new;
+       }
+       else {
+               /* Add this after the first entry for this interface */
+               bp_new->bif_next = bp_first->bif_next;
+               bp_first->bif_next = bp_new;
+       }
+       
        /*
         * Compute the length of the bpf header.  This is not necessarily
         * equal to SIZEOF_BPF_HDR because we want to insert spacing such
         * that the network layer header begins on a longword boundary (for
         * performance reasons and to alleviate alignment restrictions).
         */
-       bp->bif_hdrlen = BPF_WORDALIGN(hdrlen + SIZEOF_BPF_HDR) - hdrlen;
+       bp_new->bif_hdrlen = BPF_WORDALIGN(hdrlen + SIZEOF_BPF_HDR) - hdrlen;
+       
+       /* Take a reference on the interface */
+       ifnet_reference(ifp);
+
+       lck_mtx_unlock(bpf_mlock);
 
 #ifndef __APPLE__
        if (bootverbose)
                printf("bpf: %s%d attached\n", ifp->if_name, ifp->if_unit);
 #endif
+
+       return 0;
 }
 
 /*
@@ -1589,76 +2201,91 @@ bpfattach(ifp, dlt, hdrlen)
  * ENXIO.
  */
 void
-bpfdetach(ifp)
-       struct ifnet *ifp;
+bpfdetach(struct ifnet *ifp)
 {
-       struct bpf_if   *bp, *bp_prev;
+       struct bpf_if   *bp, *bp_prev, *bp_next;
+       struct bpf_if   *bp_free = NULL;
        struct bpf_d    *d;
-       int     s;
 
-       s = splimp();
+       
+       lck_mtx_lock(bpf_mlock);
 
        /* Locate BPF interface information */
        bp_prev = NULL;
-       for (bp = bpf_iflist; bp != NULL; bp = bp->bif_next) {
-               if (ifp == bp->bif_ifp)
-                       break;
-               bp_prev = bp;
-       }
-
-#ifdef __APPLE__
-       /* Check for no BPF interface information */
-       if (bp == NULL) {
-               return;
-       }
-#endif
-
-       /* Interface wasn't attached */
-       if (bp->bif_ifp == NULL) {
-               splx(s);
-#ifndef __APPLE__
-               printf("bpfdetach: %s%d was not attached\n", ifp->if_name,
-                   ifp->if_unit);
-#endif
-               return;
-       }
-
-       while ((d = bp->bif_dlist) != NULL) {
-               bpf_detachd(d);
-               bpf_wakeup(d);
+       for (bp = bpf_iflist; bp != NULL; bp = bp_next) {
+               bp_next = bp->bif_next;
+               if (ifp != bp->bif_ifp) {
+                       bp_prev = bp;
+                       continue;
+               }
+               
+               while ((d = bp->bif_dlist) != NULL) {
+                       bpf_detachd(d);
+                       bpf_wakeup(d);
+               }
+       
+               if (bp_prev) {
+                       bp_prev->bif_next = bp->bif_next;
+               } else {
+                       bpf_iflist = bp->bif_next;
+               }
+               
+               bp->bif_next = bp_free;
+               bp_free = bp;
+               
+               ifnet_release(ifp);
        }
 
-       if (bp_prev) {
-               bp_prev->bif_next = bp->bif_next;
-       } else {
-               bpf_iflist = bp->bif_next;
-       }
+       lck_mtx_unlock(bpf_mlock);
 
        FREE(bp, M_DEVBUF);
 
-       splx(s);
 }
 
 void
-bpf_init(unused)
-       void *unused;
+bpf_init(__unused void *unused)
 {
 #ifdef __APPLE__
        int     i;
        int     maj;
 
-       if (!bpf_devsw_installed ) {
+       if (bpf_devsw_installed == 0) {
                bpf_devsw_installed = 1;
+
+        bpf_mlock_grp_attr = lck_grp_attr_alloc_init();
+
+        bpf_mlock_grp = lck_grp_alloc_init("bpf", bpf_mlock_grp_attr);
+
+        bpf_mlock_attr = lck_attr_alloc_init();
+
+        bpf_mlock = lck_mtx_alloc_init(bpf_mlock_grp, bpf_mlock_attr);
+
+               if (bpf_mlock == 0) {
+                       printf("bpf_init: failed to allocate bpf_mlock\n");
+                       bpf_devsw_installed = 0;
+                       return;
+               }
+               
                maj = cdevsw_add(CDEV_MAJOR, &bpf_cdevsw);
                if (maj == -1) {
+                       if (bpf_mlock)
+                               lck_mtx_free(bpf_mlock, bpf_mlock_grp);
+                       if (bpf_mlock_attr)
+                               lck_attr_free(bpf_mlock_attr);
+                       if (bpf_mlock_grp)
+                               lck_grp_free(bpf_mlock_grp);
+                       if (bpf_mlock_grp_attr)
+                               lck_grp_attr_free(bpf_mlock_grp_attr);
+                       
+                       bpf_mlock = NULL;
+                       bpf_mlock_attr = NULL;
+                       bpf_mlock_grp = NULL;
+                       bpf_mlock_grp_attr = NULL;
+                       bpf_devsw_installed = 0;
                        printf("bpf_init: failed to allocate a major number!\n");
-                       nbpfilter = 0;
-                       return;
-               }
-               if (bpf_dtab_grow(NBPFILTER) == 0) {
-                       printf("bpf_init: failed to allocate bpf_dtab\n");
                        return;
                }
+
                for (i = 0 ; i < NBPFILTER; i++)
                        bpf_make_dev_t(maj);
        }
@@ -1671,51 +2298,18 @@ bpf_init(unused)
 SYSINIT(bpfdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE+CDEV_MAJOR,bpf_drvinit,NULL)
 #endif
 
-#else /* !BPF */
-#ifndef __APPLE__
-/*
- * NOP stubs to allow bpf-using drivers to load and function.
- *
- * A 'better' implementation would allow the core bpf functionality
- * to be loaded at runtime.
- */
-
-void
-bpf_tap(ifp, pkt, pktlen)
-       struct ifnet *ifp;
-       register u_char *pkt;
-       register u_int pktlen;
+#if CONFIG_MACF_NET
+struct label *
+mac_bpfdesc_label_get(struct bpf_d *d)
 {
-}
 
-void
-bpf_mtap(ifp, m)
-       struct ifnet *ifp;
-       struct mbuf *m;
-{
+       return (d->bd_label);
 }
 
 void
-bpfattach(ifp, dlt, hdrlen)
-       struct ifnet *ifp;
-       u_int dlt, hdrlen;
+mac_bpfdesc_label_set(struct bpf_d *d, struct label *label)
 {
-}
 
-void
-bpfdetach(ifp)
-       struct ifnet *ifp;
-{
+       d->bd_label = label;
 }
-
-u_int
-bpf_filter(pc, p, wirelen, buflen)
-       register const struct bpf_insn *pc;
-       register u_char *p;
-       u_int wirelen;
-       register u_int buflen;
-{
-       return -1;      /* "no filter" behaviour */
-}
-#endif /* !defined(__APPLE__) */
-#endif /* NBPFILTER > 0 */
+#endif